dsh-screenshot-plugin

When chatting with an agent in DeepSeek Harness Desktop, you often need to grab a specific region of your screen and hand it over for visual analysis—UI layout, error dialogs, form state. The old flow means alt-tabbing to a system snipping tool, saving the file, tabbing back, and typing the path by hand. dsh-screenshot-plugin collapses that entire round-trip into a single ✂ button that lives right next to the composer.

Click the button and the screen dims; drag to draw a selection rectangle. After you release, the selection stays live—you can reposition it by dragging the body and fine-tune the size with eight handles (four corners, four edges) before committing. Double-click or press Enter to capture, Esc to cancel. The PNG lands in your designated folder with an auto-numbered filename (Screenshot-N_HHmm.png), and a plain-ASCII marker such as [Shot 3 14:22] drops straight into the input box. Send that marker and a vision-capable agent like ModLens can locate and read the file without you ever naming a path.

This plugin is aimed at Windows users on DSH Desktop who keep repeating the screenshot-then-tell-the-agent loop—frontend debugging, UI comparison, error triage, form-state capture. The selection overlay is rendered with PowerShell WinForms in pure-ASCII source, avoiding encoding pitfalls in file names and cross-process hand-off; the marker itself is ASCII so it survives chat transport intact. If that grab-and-point step shows up a few times a week in your workflow, this plugin eliminates the context-switching overhead each time.